This paper details a comprehensive study benchmarking 40 computational methods designed for integrating single-cell multimodal omics data. The authors categorize these methods into four prototypical types, such as vertical and diagonal integration, based on how they handle different input data structures. The performance of these tools is systematically evaluated across seven key tasks, including dimension reduction, batch correction, and classification, using extensive real and simulated datasets. Results indicate that the effectiveness of any given method is highly dependent on the specific task, the combination of data modalities (like RNA, ADT, and ATAC), and the chosen evaluation metrics. For instance, the study noted a frequent trade-off between preserving biological signals and achieving effective batch harmonization. Ultimately, this research provides a vital guideline and recommendations for the most appropriate method selection for various analysis goals in the rapidly evolving field of single-cell omics.

References:

• Liu C, Ding S, Kim H J, et al. Multitask benchmarking of single-cell multimodal omics integration methods[J]. Nature Methods, 2025: 1-12.